Home / All Skills / Content & Media Creation / skill-creator
English (EN)简体中文 (ZH)日本語 (JA)
Content & Media Creation · freestylefly/canghe-skills

skill-creator

Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.

L4 High-Privilege · Graded auth & full auditV1 Commit-PinnedExecute CommandWrite FileDual-pass Audit
Author freestyleflyGitHub freestylefly/canghe-skills ↗License MIT（仓库根 LICENSE）Commit dd0bf35595
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

Meta-skill: A guide on "how to create a skill" in itself, with its core deliverables being a set of process specifications + three local Python tools.

skills/skill-creator/SKILL.md
Skill creation involves these steps:
What is being done here: Breaking down skill creation into a 6-step actionable workflow.

Primary principle is "conciseness": The context window is a shared resource; only provide information that the model does not already know.

skills/skill-creator/SKILL.md
The context window is a public good.
What is being done here: Elevating token cost into a design constraint, requiring questioning the necessity of every single sentence.

Select degrees of freedom based on task fragility: Use text instructions (high freedom) for variable scenarios, and parameterized scripts (low freedom) for fragile workflows.

skills/skill-creator/SKILL.md
**High freedom (text-based instructions)**: Use when multiple approaches are valid, decisions depend on context, or heuristics guide the approach.
What is being done here: Providing the analogy of "add guardrails on narrow bridges, let roam freely in open fields" to determine implementation form.

Specifies skill directory anatomy: Required SKILL.md (frontmatter + body), optional scripts/references/assets resource directories.

skills/skill-creator/SKILL.md
Every skill consists of a required SKILL.md file and optional bundled resources:
What is being done here: Defining the standard directory structure, with a section explaining the purpose of each of the three resource types.

Progressive disclosure with three-level loading: Metadata is always in context, the body loads upon trigger, and resources load on demand.

skills/skill-creator/SKILL.md
Skills use a three-level loading system to manage context efficiently:
What is being done here: Explaining why implementation details should be moved out of SKILL.md.

Step 3 uses init_skill.py to generate the template directory (SKILL.md + sample files in scripts/references/assets).

skills/skill-creator/SKILL.md
scripts/init_skill.py <skill-name> --path <output-directory>
What is being done here: Scaffolding command that executes the bundled script.

Step 5 uses package_skill.py to validate before packaging into a .skill (zip format) file.

skills/skill-creator/SKILL.md
scripts/package_skill.py <path/to/skill-folder>
What is being done here: Delivery stage, automatically running validation before packaging.

Explicitly forbids including miscellaneous docs like README/INSTALLATION_GUIDE/CHANGELOG to prevent context pollution.

skills/skill-creator/SKILL.md
Do NOT create extraneous documentation or auxiliary files, including:
What is being done here: Formulating "what not to write" into hard rules as well.

2Core Capabilities

01Skill directory structure specification: SKILL.md (frontmatter + body) + scripts/references/assets
02Six-step creation workflow: Understand → Plan → Initialize → Edit → Package → Iterate
03Scaffold generation: init_skill.py creates directories, writes the SKILL.md template + three sample resources
04Validator: quick_validate.py checks frontmatter field allowlist, naming conventions, and description length
05Packaging and distribution: package_skill.py packages into .skill (zip) upon passing validation
06Design pattern references: output-patterns.md (templates/examples) and workflows.md (sequential/conditional workflows)
07Progressive disclosure guidance: Keep SKILL.md body under 500 lines, moving details out to reference files
08Degree of freedom guidance: Choose between text instructions / parameterized scripts / fixed scripts based on task fragility

3External Dependencies

KindDependency / Endpoint
clipython3
packagePyYAML（解析 frontmatter）

4Recommended Governance & Evidence

Suggested Controls for Safe Operation
For Agent Runtime
Graded/per-use authorization, sandbox recommended
For Administrators
Approval workflow, rollback support, full audit
Evidence & Triggers (File Paths & Excerpts)
  • scripts/: Execute local script or command: scripts/init_skill.py
  • 本地工作目录: File read/write and artifact output: Create skill directory and SKILL.md
Audit Notes: Evaluated as 🔵 blue: The skill bundles three local Python scripts that perform local read/write operations such as directory creation, template file generation, and zip packaging (§3 disk writing matches init_skill.py/package_skill.py across 10 rules). §3 shows no network requests, no credential reading, and no danger flags. Risk reminder: blue, standard usage; scripts write to the file system according to paths provided by the user, so pay attention to the target directory.

5Second-Pass Verification

  • [ok] 文件面核对 — §1 共 7 个文件：SKILL.md、LICENSE.txt、2 个 references、3 个 scripts，与 internal_assets 一致。
  • [ok] 脚本执行真伪 — §3「执行外部程序」无命中（脚本由模型按 SKILL.md 指示调用）；三个脚本源码实读确认存在。
  • [ok] 凭证读取反例 — §3 两条凭证命中经回原文核对为 SKILL.md 的 'challenge each' 文本与示例 'token validation middleware'，非 env/凭据读取，故 credential_reads 判空。
  • [ok] 写盘复核 — §3 写盘 10 条命中 init_skill.py（mkdir/write_text/chmod）与 package_skill.py（mkdir/zipf.write），与 file_writes 一致。
  • [ok] 网络与注入面 — §3 无 URL、无网络请求命中；注入面仅剩用户输入经 yaml.safe_load 解析的语义校验缺口（设计边界）。
  • [ok] 档位判定复核 — 本地脚本 + 本地读写、无网络无凭证 → 🔵 blue 成立。

6Verdict & Scenarios

  • Solidifies skill design experience into actionable workflows and principles (conciseness / degrees of freedom / progressive disclosure).
  • The trio of scaffolding/validation/packaging scripts forms a closed loop, reducing manual errors.
  • Clear target structures and anti-patterns (e.g., banning README/CHANGELOG).
    • Best for: Authors aiming to create or iterate on agent skills: Those needing standardized directory structures, frontmatter specifications, and workflow support via scaffolding and packaging tools.
    Not for: Scenarios where you only want to write a quick prompt without packaging/distribution; cases requiring strict security audits (script content review) before publishing; non-skills-compatible hosts.
    Install Agent-Ready
    ① Verified Mirror 2026-10-10
    Option A · Direct DownloadDownload skill-creator.tar.gz
    sha256: 0b2639ac219971bd…
    Option B · Copy Guide for Agent
    install.json
    Agents automatically download and verify mirrors from this JSON guide.
    ② Upstream GitHub
    GitHub Repo ↗
    Pinned to commit dd0bf35595.